Training my own dataset

[charliewang789]

Hi
what would be the data format and folder layout that I need to setup if I have my own dataset?

Thx

[jgcbrouns]

+1

[Wangyy-moya]

the same question

[aniketgokhale]

Here's the solution that worked for me.

I created a csv file containing path for image and its label.
I wrote a custom class for data loading.

import os
import pandas as pd
import torch
from torch.utils.data import Dataset
from skimage import io
from skimage.transform import resize


class SignatureDataset(Dataset):
    def __init__(self,csv_file, root_dir, transform = None):
        self.annotations = pd.read_csv(csv_file)
        self.root_dir = root_dir
        self.transform = transform

    def __len__(self):
        return len(self.annotations)

    def __getitem__(self,index):
        img_path = os.path.join(self.root_dir, self.annotations.iloc[index,0])
        image =  io.imread(img_path)
        image = resize(image,(3,256,256))
        y_label = torch.tensor(int(self.annotations.iloc[index,1]))

        if self.transform:
            image = self.transform(image)

        return (torch.from_numpy(image.copy()).float(),y_label)

I changed bit_hyperrule.py accordingly.
Here are the changes:

known_dataset_sizes = {
  'cifar10': (32, 32),
  'cifar100': (32, 32),
  'oxford_iiit_pet': (224, 224),
  'oxford_flowers102': (224, 224),
  'imagenet2012': (224, 224),
  'customDataset' : (256,256),
}

You can change the resolution from the get_resolution function (from bit.hyperrule.py) if needed.

I changed /bit_pytorch/train.py file to load the custom data.

  elif args.dataset == "imagenet2012":
    train_set = tv.datasets.ImageFolder(pjoin(args.datadir, "train"), train_tx)
    valid_set = tv.datasets.ImageFolder(pjoin(args.datadir, "val"), val_tx)
  elif args.dataset == "customDataset":
    dataset = customDataset.SignatureDataset(csv_file="customDataLoader.csv", root_dir="path to custom dataset", transform = None)
    train_set, valid_set = torch.utils.data.random_split(dataset, [10841,2711])     #[training, validation]

Keep customDataLoader.csv as well as customDataset.py in the root folder (with bit_hyperrule.py).

Run the code using command:
python -m bit_pytorch.train --name custom_classifier --model BiT-M-R50x1 --logdir /tmp/bit_logs --dataset customDataset.

I had changed the default values (for batch_size, etc.) from the code itself.

Hope that helps. !

References:
To create customDataset.py, you can refer to https://www.youtube.com/watch?v=ZoZHd0Zm3RY

[Wangyy-moya]

Here's the solution that worked for me.

I created a csv file containing path for image and its label.
I wrote a custom class for data loading.

import os
import pandas as pd
import torch
from torch.utils.data import Dataset
from skimage import io
from skimage.transform import resize


class SignatureDataset(Dataset):
    def __init__(self,csv_file, root_dir, transform = None):
        self.annotations = pd.read_csv(csv_file)
        self.root_dir = root_dir
        self.transform = transform

    def __len__(self):
        return len(self.annotations)

    def __getitem__(self,index):
        img_path = os.path.join(self.root_dir, self.annotations.iloc[index,0])
        image =  io.imread(img_path)
        image = resize(image,(3,256,256))
        y_label = torch.tensor(int(self.annotations.iloc[index,1]))

        if self.transform:
            image = self.transform(image)

        return (torch.from_numpy(image.copy()).float(),y_label)

I changed bit_hyperrule.py accordingly.
Here are the changes:

known_dataset_sizes = {
  'cifar10': (32, 32),
  'cifar100': (32, 32),
  'oxford_iiit_pet': (224, 224),
  'oxford_flowers102': (224, 224),
  'imagenet2012': (224, 224),
  'customDataset' : (256,256),
}

You can change the resolution from the get_resolution function (from bit.hyperrule.py) if needed.

I changed /bit_pytorch/train.py file to load the custom data.

  elif args.dataset == "imagenet2012":
    train_set = tv.datasets.ImageFolder(pjoin(args.datadir, "train"), train_tx)
    valid_set = tv.datasets.ImageFolder(pjoin(args.datadir, "val"), val_tx)
  elif args.dataset == "customDataset":
    dataset = customDataset.SignatureDataset(csv_file="customDataLoader.csv", root_dir="path to custom dataset", transform = None)
    train_set, valid_set = torch.utils.data.random_split(dataset, [10841,2711])     #[training, validation]

Keep customDataLoader.csv as well as customDataset.py in the root folder (with bit_hyperrule.py).

Run the code using command:
python -m bit_pytorch.train --name custom_classifier --model BiT-M-R50x1 --logdir /tmp/bit_logs --dataset customDataset.

I had changed the default values (for batch_size, etc.) from the code itself.

Hope that helps. !

References:
To create customDataset.py, you can refer to https://www.youtube.com/watch?v=ZoZHd0Zm3RY

I have tried yours,but it raise:AttributeError: 'MyDataset' object has no attribute 'classes'.

[aniketgokhale]

Here's the solution that worked for me.
I created a csv file containing path for image and its label.
I wrote a custom class for data loading.

import os
import pandas as pd
import torch
from torch.utils.data import Dataset
from skimage import io
from skimage.transform import resize


class SignatureDataset(Dataset):
    def __init__(self,csv_file, root_dir, transform = None):
        self.annotations = pd.read_csv(csv_file)
        self.root_dir = root_dir
        self.transform = transform

    def __len__(self):
        return len(self.annotations)

    def __getitem__(self,index):
        img_path = os.path.join(self.root_dir, self.annotations.iloc[index,0])
        image =  io.imread(img_path)
        image = resize(image,(3,256,256))
        y_label = torch.tensor(int(self.annotations.iloc[index,1]))

        if self.transform:
            image = self.transform(image)

        return (torch.from_numpy(image.copy()).float(),y_label)

I changed bit_hyperrule.py accordingly.
Here are the changes:

known_dataset_sizes = {
  'cifar10': (32, 32),
  'cifar100': (32, 32),
  'oxford_iiit_pet': (224, 224),
  'oxford_flowers102': (224, 224),
  'imagenet2012': (224, 224),
  'customDataset' : (256,256),
}

You can change the resolution from the get_resolution function (from bit.hyperrule.py) if needed.
I changed /bit_pytorch/train.py file to load the custom data.

  elif args.dataset == "imagenet2012":
    train_set = tv.datasets.ImageFolder(pjoin(args.datadir, "train"), train_tx)
    valid_set = tv.datasets.ImageFolder(pjoin(args.datadir, "val"), val_tx)
  elif args.dataset == "customDataset":
    dataset = customDataset.SignatureDataset(csv_file="customDataLoader.csv", root_dir="path to custom dataset", transform = None)
    train_set, valid_set = torch.utils.data.random_split(dataset, [10841,2711])     #[training, validation]

Keep customDataLoader.csv as well as customDataset.py in the root folder (with bit_hyperrule.py).
Run the code using command:
python -m bit_pytorch.train --name custom_classifier --model BiT-M-R50x1 --logdir /tmp/bit_logs --dataset customDataset.
I had changed the default values (for batch_size, etc.) from the code itself.
Hope that helps. !
References:
To create customDataset.py, you can refer to https://www.youtube.com/watch?v=ZoZHd0Zm3RY

I have tried yours,but it raise:AttributeError: 'MyDataset' object has no attribute 'classes'.

When I tried the mentioned solution, it worked.
Can you try changing line no. 205 to:

model = models.KNOWN_MODELS[args.model](head_size=len(valid_set[1]), zero_head=True)

[Wangyy-moya]

Here's the solution that worked for me.
I created a csv file containing path for image and its label.
I wrote a custom class for data loading.

import os
import pandas as pd
import torch
from torch.utils.data import Dataset
from skimage import io
from skimage.transform import resize


class SignatureDataset(Dataset):
    def __init__(self,csv_file, root_dir, transform = None):
        self.annotations = pd.read_csv(csv_file)
        self.root_dir = root_dir
        self.transform = transform

    def __len__(self):
        return len(self.annotations)

    def __getitem__(self,index):
        img_path = os.path.join(self.root_dir, self.annotations.iloc[index,0])
        image =  io.imread(img_path)
        image = resize(image,(3,256,256))
        y_label = torch.tensor(int(self.annotations.iloc[index,1]))

        if self.transform:
            image = self.transform(image)

        return (torch.from_numpy(image.copy()).float(),y_label)

I changed bit_hyperrule.py accordingly.
Here are the changes:

known_dataset_sizes = {
  'cifar10': (32, 32),
  'cifar100': (32, 32),
  'oxford_iiit_pet': (224, 224),
  'oxford_flowers102': (224, 224),
  'imagenet2012': (224, 224),
  'customDataset' : (256,256),
}

You can change the resolution from the get_resolution function (from bit.hyperrule.py) if needed.
I changed /bit_pytorch/train.py file to load the custom data.

  elif args.dataset == "imagenet2012":
    train_set = tv.datasets.ImageFolder(pjoin(args.datadir, "train"), train_tx)
    valid_set = tv.datasets.ImageFolder(pjoin(args.datadir, "val"), val_tx)
  elif args.dataset == "customDataset":
    dataset = customDataset.SignatureDataset(csv_file="customDataLoader.csv", root_dir="path to custom dataset", transform = None)
    train_set, valid_set = torch.utils.data.random_split(dataset, [10841,2711])     #[training, validation]

Keep customDataLoader.csv as well as customDataset.py in the root folder (with bit_hyperrule.py).
Run the code using command:
python -m bit_pytorch.train --name custom_classifier --model BiT-M-R50x1 --logdir /tmp/bit_logs --dataset customDataset.
I had changed the default values (for batch_size, etc.) from the code itself.
Hope that helps. !
References:
To create customDataset.py, you can refer to https://www.youtube.com/watch?v=ZoZHd0Zm3RY

I have tried yours,but it raise:AttributeError: 'MyDataset' object has no attribute 'classes'.

When I tried the mentioned solution, it worked.
Can you try changing line no. 205 to:

model = models.KNOWN_MODELS[args.model](head_size=len(valid_set[1]), zero_head=True)

OK,Thanks